LetterMint AI – Browser-Native AI Content Writer React App
LetterMint AI – The Zero-Cost, Browser-Native AI Content Platform
LetterMint AI is a complete AI-powered content creation and productivity suite that runs full Large Language Models entirely inside the user’s browser. No backend servers, no API keys, no monthly subscription fees. Powered by WebLLM and WebGPU hardware acceleration, it delivers real AI inference using models like Llama 3.2, Phi 3.5, and Gemma 2 on hardware users already own.
This is not a wrapper around a cloud API. This is not a UI mockup with fake responses. LetterMint AI downloads real LLM weights into the browser cache and runs genuine on-device inference with streaming output. Completely offline-capable once models are cached.
Why LetterMint AI is Different
- Zero API costs, forever. Models run locally via WebGPU. You pay $0.00 for inference, no matter how many users you scale to. No OpenAI bills, no Anthropic invoices, no token metering.
- Absolute privacy. No data ever leaves the device. No telemetry, no cloud processing. Ideal for enterprise compliance, legal writing, medical notes, or personal journaling.
- No backend required. Pure front-end architecture built with React + Vite. Deploy to any static host (Vercel, Netlify, Firebase Hosting, GitHub Pages) at zero cost.
- Real AI, not a demo. On supported browsers (Chrome 113+ / Edge 113+), the AI Writer, Summarizer, Code Generator, and all 13 tool categories produce genuine model-generated output, not pre-written placeholder text.
Core Features
- AI Writer and Editor. Rich text editing with inline AI generation, rewriting, grammar correction, and tone adjustment. Supports markdown and real-time streaming output from the local model.
- 13 AI-Powered Tools. SQL Generator, Code Writer, Stripe API Generator, Python Generator, Summarizer, Translator, Regex Builder, JSON Formatter, Text Classifier, Data Extractor, and more. Each tool uses dedicated system prompts for accurate, context-aware generation.
- Knowledge Base and RAG. Upload local documents and ask the AI questions grounded in your content. Real retrieval-augmented generation running entirely in-browser.
- 30+ Writing Templates. Blog posts, marketing emails, social media copy, technical documentation, sales pitches, cover letters, and more. Each template includes a tuned system prompt that drives actual AI output. Users can also create and save custom templates.
- Document and Project Management. Create, organize, tag, search, and manage documents and projects. Full CRUD with persistent storage across browser sessions.
- Export. Download documents as PDF, Markdown, or plain text. Fully client-side, no server processing required.
- Dashboard with Activity Tracking. Documents created, words generated, time saved, and AI requests. Tracked from real user activity and updated live.
- Global Search (Ctrl+K). Instant search across all documents, tools, templates, and pages.
- Model Selection and Configuration. Choose between Llama 3.2 (1B / 3B), Phi 3.5 Mini, or Gemma 2 (2B). Adjust temperature, token limits, and generation parameters per session.
- Light and Dark Mode. Fully integrated theme switching with smooth transitions.
- Premium UI/UX. Built with Tailwind CSS, Framer Motion animations, and Shadcn UI components for a polished, responsive, production-grade interface.
Built for Developers
Clean, modern, and highly extensible codebase designed for easy customization:
| Layer | Technology |
|---|---|
| Framework | React 19 + Vite 6 |
| Styling | Tailwind CSS + Radix UI Primitives + Shadcn UI |
| AI Engine | @mlc-ai/web-llm (WebGPU accelerated) |
| Routing | React Router v7 |
| Storage | IndexedDB for model caching, localStorage for app data |
| Animations | Framer Motion |
| Testing | Vitest |
What’s Fully Working vs. Demo Scaffolding
LetterMint AI is a working product, not a static mockup. The table below clearly separates genuinely functional features from the lightweight scaffolding included so the template runs standalone without a backend.
Fully Working Features
| Feature | Details |
|---|---|
| AI Text Generation | Real on-device LLM inference via WebLLM + WebGPU. Models are downloaded to the browser cache and run locally with streaming output. This is genuine AI, not pre-written text. |
| AI Writer and All AI Tools | All 13 tool categories: Writer, Summarizer, SQL, Code, Stripe API, Python, Translation, Regex, JSON, Classification, Extraction, and more. All powered by real local model inference. |
| Rich Text Document Editor | Full-featured editor with formatting toolbar, markdown support, and real-time editing. Documents are created, edited, and saved. |
| Document and Project Management | Create, organize, search, and manage documents and projects. All CRUD operations work with persistent storage. |
| Knowledge Base / RAG | Upload documents and ask AI questions about them. Real in-browser document Q&A powered by the local model. |
| Export (PDF, Markdown, Text) | Fully working client-side export. Documents are converted and downloaded with no server involved. |
| 30+ AI Writing Templates | Each template includes real system prompts that drive actual AI generation. Users can create and save custom templates. |
| Dashboard Activity Tracking | Stats are tracked from real user activity and update as the app is used. |
| Settings and Model Configuration | Model selection, generation parameters, and workspace preferences. All functional and persisted. |
| Global Search (Ctrl+K) | Search across documents, tools, and pages. Fully functional. |
AI Fallback Mode
When WebGPU is not supported (older browsers, unsupported GPUs, insufficient VRAM), the app automatically switches to Fallback Mode:
- All AI tools remain usable. They return sample/placeholder responses so users can fully explore the interface and workflows.
- The UI clearly labels this state as “Fallback Mode (Mock Responses)” so there is no ambiguity.
- Fallback covers all 13 AI tool categories.
- This ensures the live demo works for all reviewers and buyers, regardless of their hardware.
On any device with Chrome 113+ or Edge 113+, Fallback Mode is bypassed entirely and the AI runs real local model inference.
Demo Scaffolding (Easily Replaceable)
These peripheral features use lightweight client-side simulation so the template works out of the box without a backend. Each is designed to be swapped with your own integration in minutes:
| Feature | Current Behavior | How to Replace |
|---|---|---|
| Authentication | Login/Register forms are fully built with validation. On login, a demo user profile is created in localStorage. No real backend auth. | Integrate Firebase Auth, Supabase Auth, Clerk, or any provider. The AuthContext is already abstracted for easy replacement. |
| Profile Save | Profile updates save to localStorage with a realistic delay. Password change and avatar buttons are UI-ready but not connected. | Connect to your user API endpoint. |
| Data Persistence | All documents, projects, templates, and usage stats are stored in browser localStorage. Persistent across sessions but local to the device. | Replace localStorage calls with Firebase Firestore, Supabase, MongoDB, or any database. |
| Contact Form | Fully designed form on the Support page. Requires EmailJS credentials in the .env file to send real emails. | Add your EmailJS keys or replace with your own email service. |
| Dashboard Trend Indicators | The percentage values (+12%, +23%) next to dashboard stats are static/decorative. | Calculate from historical data once a real database is connected. |
| Upgrade Plan Button | UI-only button on the Profile page. | Connect to Stripe, Paddle, or your billing provider. |
In short: The AI engine, document editor, all 13 AI tools, templates, export, search, knowledge base, and dashboard tracking are genuinely working features. The scaffolded parts (auth, profile save, data storage) are standard template conveniences that every UI template includes. Designed to be replaced in minutes with your preferred backend.
Important Requirement: LetterMint AI runs real hardware-accelerated LLM inference. It requires a modern browser with WebGPU support enabled (Google Chrome 113+, Microsoft Edge 113+, or equivalent). On unsupported browsers, the app functions normally in Fallback Mode with sample AI responses.
What’s Included
- Complete React/Vite source code
- Comprehensive installation and configuration documentation
- Production-ready build scripts
